Numerical simulations have gained primary importance for mechanical design over the last decades. However, a simulation must be re-run each time an engineer wishes to change any aspect of the object being designed, leading to a slow and costly engineering process. Direct applications that would be enabled by faster simulations are extremely wide, including interactive design exploration, real-time control in digital twins, and large-scale shape optimization. In this talk, we explain how a new generation of software developed by Neural Concept, powered by recent algorithms based on geometric deep-learning, allows shortcutting any simulation chain through a predictive model that outputs post processed results directly from the CAD design. These models are being used in engineering companies to simplify processes and to emulate the expertise of simulation engineers in the hands of product or design engineers early in the development process. Thus, the number of iterations between teams is reduced, while the design activities are accelerated. In traditional approaches, models are trained on a specific parametric representation of the design space. Any modification in the design space or boundary conditions requires generating a new dataset of simulations and training an independent model. In many industry sectors – from the automotive industry to building design – each project involves many changes in the description of the parametric space or in external conditions. The necessity to generate a completely new dataset at each iteration, without being able to leverage simulations collected during previous iterations can be frustrating for the engineer. NC Shape, our software, can be used to build surrogate models of numerical solvers, while also being agnostic to the shape parameters as it directly processes the mesh representation of the design. Hence, a single predictor can be trained with a large amount of data and can be used for numerous optimization tasks and the engineer does not have to choose and stick to a specific parametrization from the beginning to the end of experiments. Furthermore, it can leverage on transfer learning abilities of deep models to blend simulations from multiple sources and with multiple fidelities. Developing new buildings for cities requires designers to consider not only how the environment around the building will affect its structure, but also how the building itself will affect the environment around it. This includes evaluating the structural integrity, wind effects on the building, sunlight exposure, and pedestrian comfort. Traditional tools used by designers to quantify a building's impact on (and from) the flow – namely wind tunnel experiments and CFD simulations – are typically slow and expensive, making them unsuitable for rapid iteration and often too costly for smaller projects. Understanding at the outset how a proposed building will interact with its surroundings, specifically pedestrian level winds, saves time and cost in catching issues that may arise in later stages of the design process. Orbital Stack, part of RWDI Ventures, works closely with Neural Concept using NC Shape for wind flow prediction on city geometries. The quality of the dataset used for training is a critical component in getting a good predictive model, and this is enabled by RWDI world’s largest urban wind flow dataset. The model was integrated within the existing Orbital Stack web-app, and currently provides any user with AI-enabled instant feedback. Neural Concept developed a robust pipeline to accept user cities in the dataset and process them in a solid and reliable way, compatible with any innovations the designers include to the city, and with a short processing time. The cities in the dataset have wide variability in geometry (size, building styles, level of detail), and the resulting flow, complex by nature, drastically changes between designs. NC Shape's ability to learn different categories of designs, leverage large databases of historical simulations, and predict complex relationships, lead to a good match compared to the classical simulations. NC Shape also outperforms other commercially available AI solutions and is especially better for non-standard use cases: varying topology or when a parametrization of the shape is not available, something that other commercially available solutions are not able to handle. This tool allows building designers to compute approximate solutions orders of magnitude faster than the typical numerical simulators, in order to interact in real-time with design iterations. Beyond that, it allows to compute the gradients of an objective function with respect to the input shape. This ultimately permits the user to run optimization loops using first-order optimization methods. It allows faster design cycles, providing on-demand, reliable airflow analysis, increased performance and sustainability of the final design.